Clausal coordination and coordinate ellipsis in a model of ...
Transcript of Clausal coordination and coordinate ellipsis in a model of ...
Clausal coordination and coordinate ellipsisin a model of the speaker*
GERARD KEMPEN
Abstract
This article presents a psycholinguistically inspired approach to the syntax
of clause-level coordination and coordinate ellipsis. It departs from the as-
sumption that coordinations are structurally similar to so-called appropri-
ateness repairs — an important type of self-repairs in spontaneous speech.
Coordinate structures and appropriateness repairs can both be viewed as
‘‘update’’ constructions. Updating is defined as a special sentence produc-
tion mode that e‰ciently revises or augments existing sentential structure
in response to modifications in the speaker’s communicative intention. This
perspective is shown to o¤er an empirically satisfactory and theoretically
parsimonious account of two prominent types of coordinate ellipsis, in par-
ticular ‘‘forward conjunction reduction’’ (FCR) and ‘‘gapping’’ (including
‘‘long-distance gapping’’ and ‘‘subgapping’’). They are analyzed as di¤er-
ent manifestations of ‘‘incremental updating’’ — e‰cient updating of only
part of the existing sentential structure. Based on empirical data from
Dutch and German, novel treatments are proposed for both types of clausal
coordinate ellipsis.
The coordination-as-updating perspective appears to explain some gen-
eral properties of coordinate structure: the existence of the well-known ‘‘co-
ordinate structure constraint’’, and the attractiveness of three-dimensional
representations of coordination. Moreover, two other forms of coordinate
ellipsis — SGF (‘‘subject gap in finite clauses with fronted verb’’), and
‘‘backward conjunction reduction’’ (BCR) (also known as ‘‘right node rais-
ing’’ or RNR) — are shown to be incompatible with the notion of incre-
mental updating. Alternative theoretical interpretations of these phenomena
are proposed.
The four types of clausal coordinate ellipsis — SGF, gapping, FCR and
BCR — are argued to originate in four di¤erent stages of sentence produc-
tion: Intending (i.e., preparing the communicative intention), conceptual-
ization, grammatical encoding, and phonological encoding, respectively.
Linguistics 47–3 (2009), 653–696
DOI 10.1515/LING.2009.022
0024–3949/09/0047–0653
6 Walter de Gruyter
1. Introduction
This article deals with the syntactic aspects of coordination and coordi-
nate ellipsis in clause-level coordination with the conjunction and. The
theoretical framework to be presented here proceeds from the assumption
that two key elliptical phenomena, ‘‘forward conjunction reduction’’
(FCR) and ‘‘gapping’’, result from an update process that also underliescertain types of self-repairs in spontaneous speech — in particular so-
called appropriateness repairs. These repairs occur when speakers revise
the meaning underlying all or part of what they have just said in the cur-
rent sentence and replace it by wordings that fit in with the revised mean-
ing. Often, the ensuing ‘‘update’’ (revision, edit) does not replace the
entire utterance but only a part that includes the new wordings. Coordi-
nation and self-repair thus appear as cognate processes of conceptual and
syntactic updating.1 I show that this perspective enables an empiricallysatisfactory and theoretically parsimonious analysis of FCR and gapping
in Dutch and German.
FCR and gapping are illustrated by sentences (1) and (2), respectively.
The elision in (1) targets the second conjunct — more precisely, the left
periphery of that conjunct (marked by the dots). In (2), the head verb of
the clause is elided. This elliptical variant often gives rise to elision of a
non-left-peripheral (e.g., medial) fragment.
(1) FCR: The town [S where Jan lives and . . . Piet works]
(2) Gapping: Jan lives in Leiden and Piet . . . in Nijmegen
Two other forms of coordinate ellipsis are ‘‘backward conjunction reduc-
tion’’ (BCR) (also known as ‘‘right node raising or RNR; see Postal 1974)
and SGF (‘‘subject gap in finite clauses with fronted verb’’; see Hohle
1983). BCR (3) refers to elision of a right-peripheral part of the left con-
junct. In SGF (4), the second clausal conjuncts is finite but has no overt
subject; the implied subject is identical to the overt, non-left peripheralsubject of the preceding clause. (The non-left peripherality of the overt
subject precludes an analysis in terms of FCR.)
(3) BCR: Jan lives . . . and Piet works in Leiden(4) SGF: Into the wood went the hunter and . . . shot a hare.
1.1. Preview
As will become clear shortly, the coordination-as-updating approach to
be developed in this article does not presuppose an existing declarative
654 G. Kempen
grammar formalism, nor does it embody a new declarative formalism. In-
stead, I advocate an analysis of coordination and (some of ) its elliptical
variants in procedural terms, as resulting from full or partial updates (re-
visions, edits) of structures that have already been built. Hence, it does
not serve much purpose to review theoretical treatments of coordination
in the literature extensively. (For detailed overviews, see Dik 1968; Neijt
1979; Van Oirsouw 1987; Seuren 1996; Johannessen 1998; Schwabe andZhang 2000; Haspelmath 2004; te Velde 2006.) But a survey of the key
phenomena of FCR and gapping with emphasis on their similarity to
phenomena of self-repair, is an essential ingredient. This is the topic of
Section 2. Section 3 outlines Levelt’s (1989) model of sentence production
that is presupposed in the remainder of the article. Then, in Section 4 I
sketch the place of FCR and gapping in the sentence production model.
Section 5 develops an incremental updating analysis of FCR and gapping
in monoclausal sentences and in sentences with nested clauses. In Sections6 and 7, I turn to BCR and SGF, respectively. Given the fine-structure of
these ellipsis types, I argue that both are incompatible with the notion of
coordination-as-updating (although for entirely di¤erent reasons), and
suggest directions for adequate treatments. In Section 8, I argue that the
novel perspective explains two fundamental but ill-understood properties
of gapping and FCR: why they (1) obey the ‘‘coordinate structure con-
straint’’ (CSC) with ‘‘across-the-board’’ (ATB) application of movement
operations, and (2) invite structural representations in terms of (a nota-tional variant of ) three-dimensional trees. In Section 9, finally, I point
out a number of unresolved research issues.
2. Coordination and self-repair: a comparison from the viewpoint ofsentence production
Levelt (1983, 1989: 485–489) observed a relationship between FCR and
self-repair in spontaneous speech. Such corrections usually consist of
three elements (see [5]):
– reparandum, that is, the original utterance containing an error
– editing term (. . . uh . . . , . . . I mean . . . , . . . or rather . . .), and
– repair text.
The speaker interrupts the ongoing speech, signals this to the listener bymeans of a pause and/or an editing term, retraces to an earlier point in
the utterance, and reformulates it from there. This type of self-repairs
is called retracing repairs. Levelt noticed that only certain positions in
Clausal coordination and coordinative ellipsis 655
the reparandum qualify as potential points of resumption (‘‘retracing
targets’’) and advanced a descriptive rule demarcating these positions.
The rule presupposes that the three basic elements of retracing repairs
correspond to three parts of a coordination: the lefthand (or ‘‘anterior’’)
member, the conjunction, and the righthand (or ‘‘posterior’’) member, as
shown in (5b).2 The resumption point in (5a) is the position just before
the verb. The fragment preceding this point (i.e., the subject NP) is incor-porated into the final utterance and, spliced together with the repair text,
forms a correct sentence.
(5) a. Jan koopt . . . eh steelt fietsen
Jan buys . . . uh steals bikes
reparandum editing term repair text
b. Jan koopt . . . en . . . steelt fietsen
Jan buys . . . and . . . steals bikesanterior conjunct conjunction posterior conjunct
In essence, Levelt’s rule states that a position within the reparandum
qualifies as a suitable resumption point if and only if it yields a repair
that corresponds to a grammatical coordination (FCR). For instance,
the positions marked by ‘‘!’’ in (6a) and (7a) do not qualify as potential
resumption points because the corresponding coordinations (6b) and (7b)
are ungrammatical. In fact, the only permissible resumption points arelocated before the subject NP, as shown in (6c)–(6d) and (7c)–(7d). The
symbol ‘‘¿’’ marks the position of the interrupt, i.e., the position where
the original utterance comes to a halt. (In Section 5.2.1, I will account
for ill-formedness of coordinations such as [6b] and [7b].)
The sentences in (6) and (7) also exemplify the ‘‘update diagrams’’ that
accompany many FCR and gapping cases in the sequel. The first line of
such a diagram represents the reparandum or the anterior conjunct from
the beginning of the utterance up to the interrupt. The second line showsthe repair text or the posterior conjunct. A continuous line ( )
printed below a left-peripheral string of the reparandum or anterior con-
junct indicates that this string, in e¤ect, functions as part of the repair text
or the posterior conjunct. For instance, the continuous line below is in (6c)
indicates that the meaning of the finally resulting utterance should be
based on the analysis of IS the doctor interviewing the patient rather than
the doctor interviewing the patient. The strings of asterisks (********) in
(7a) and in (7b) indicate illegal applications of the combinatorial process.In the remaining examples, I will suppress the ‘‘!’’ symbol, which always
follows immediately after the coordinator or the editing term and there-
fore is entirely redundant.
656 G. Kempen
(6) a. *Is the doctor seeing, uh, the doctor interviewing the patient?
Is the doctor seeing ¿uh ! the doctor interviewing the patient?
b. *Is the doctor seeing and the doctor interviewing the patient?
Is the doctor seeing ¿and ! the doctor interviewing the patient?
c. Is the nurse, uh, the doctor interviewing the patient?
Is the nurse ¿uh ! the doctor interviewing the patient?
d. Is the nurse or the doctor interviewing the patient?
Is the nurse ¿or ! the doctor interviewing the patient?
(Levelt 1989: 486–487)
(7) a. *He told me that this man stole bikes, uh, boy stole bikes
He told me [S that this man stole bikes******** uh ! boy stole bikes]
b. *He told me that this man stole bikes and boy stole bikes
He told me [S that this man stole bikes
******** and ! boy stole bikes]
c. He told me that this man stole bikes, uh, this boy stole bikes
He told me [S that this man stole bikes
uh ! this boy stole bikes]
d. He told me that this man stole bikes and this boy stole bikes
He told me [S that this man stole bikes
and ! this boy stole bikes]
Van Wijk and Kempen (1987) observed a second type of self-corrections
which they call substitution repairs — see (8) and (9). They crucially di¤er
from the repairs studied by Levelt in that one or more potentially non-
adjacent major constituents of the original clause, not including the head
verb, are replaced. This property also holds for the coordination analogof substitution repairs (8b) and (9b), both exemplifying gapping. The sub-
stitute constituents in the second conjunct are often called remnants. Se-
quences of dots (.............) in the second line of an update diagram show
which strings in the first line have to be combined with the remnants in
order to arrive at the correct interpretation of the repair text or the poste-
rior conjunct. Illegal applications of this combinatorial process will de-
picted by strings of asterisks (********).
(8) a. I met their eldest son yesterday, or rather, their youngest son
I met their eldest son yesterday
....... or rather their youngest son .............
Clausal coordination and coordinative ellipsis 657
b. I met their eldest son yesterday, and their youngest one
I met their eldest son yesterday
....... and their youngest son ............
(9) a. Then, the blue car crashed into the red one, I mean, the Jaguar
into the Porsche
Then, the blue car crashed into the red one
I mean, ....... the Jaguar .......... into the Porscheb. Then, the BMW crashed into the Volvo, and the Jaguar into the
Porsche
Then, the BMW crashed into the Volvo
and ........ the Jaguar ........... into the Porsche
Another similarity between substitution repairs on the one hand and gap-
ping on the other concerns the fact that the righthand conjunct cannot in-
clude an overt head verb, as illustrated in (10a). Apparently, the head
verb in a posterior clausal conjunct rules out gapping and sets the ellipsis
mode to FCR (10b).
(10) a. *As a boy he read all novels by Scott, or rather devoured
b. As a boy he read all novels by Scott, or rather devoured them
As a boy he read all novels by Scott
or rather devoured them
c. As a boy he read all novels by Scott and devoured *(them)
As a boy he read all novels by Scott
and devoured them
As suggested by the ‘‘grammatical’’ examples of self-repair above, the re-
pair text is a string of complete major constituents. Resumption points ap-
pear to be located preferably before a major constituent of a clause, not
within one. Similarly, the elision process in both gapping and FCR leavesmajor constituents intact.3
The similarities between self-repair on the one hand and FCR and gap-
ping on the other can now be summarized as follows:
– The elision in FCR is left peripheral; in self-repair, the repair text
together with all major constituents preceding the resumption point
make up the new clause.
– In gapping, a posterior conjunct may consist of major constituents
whose counterparts in the anterior (initial) conjunct are not adja-
cent; in substitution repairs, the repair text may replace nonadja-cent major constituents in the reparandum.
– The elisions in FCR and gapping respect major constituent bounda-
ries (except for long-distance gapping; see Section 5.2.2).
658 G. Kempen
– In both FCR/gapping and their self-repair analogs, the presence or
absence of an overt clausal head verb determines the continuation
options: gapping is ruled out by the presence of the overt head
verb in the posterior clause; substitution repairs only occur if the re-
pair text does not contain the overt clausal head verb.
These remarkable points of resemblance invite an account of coordina-tion and coordinate ellipsis based on their similarity with self-repairs —
in particular, with self-repairs where ‘‘appropriateness’’ is at stake. They
occur when — halfway through, or at the end of, a sentence — speakers
modify the communicatiev intention underlying the current utterance in
such a way that at least part of the utterance needs to be updated. In
such repairs, some or all of the originally intended content that already
has been encoded conceptually and grammatically4 and surfaced as an
overt utterance, is replaced by more appropriate content, which requiresat least a partially di¤erent overt realization.
An obvious objection against the suggested theoretical course follows
from the observation that in self-repairs the reparandum fragment to the
right of the resumption point is overwritten completely so that only one
constituent remains: the string consisting of the initial reparandum frag-
ment that precedes the resumption point, followed by the repair text.
This contrasts with coordinate structures, where the initial conjunct is
not superseded by a later one and at least two constituents see the lightof day. However, this observation is incorrect. While formulating the re-
pair text, the speaker presupposes that the entire reparandum forms part
of the utterance and that the listener has understood it. In variant (8c) of
(8a), for instance, the repair text contains two pronouns (their and one)
which both take a part of the reparandum (the Johnsons and son, respec-
tively) as antecedent. Hence, the part of the reparandum rightward of the
resumption point must be assumed to belong to the resulting sentence.
(8) c. I met the eldest son of the Johnsons, sorry, their youngest one,
yesterday
I met the eldest son of the Johnsons ¿sorry ! their youngest one yesterday
In descriptions of updating processes, one needs to define the target do-
main, the substitution units, and the substitution algorithm. To illustrate,
consider the situation of a software company that continually works onimprovements of its software product — some complex program — and
regularly distributes updates among its clients. If every update comprises
a complete version of the program, the ‘‘program’’ is target domain and
Clausal coordination and coordinative ellipsis 659
substitution unit at the same time. Alternatively, an update could contain
a new version of only the modified functions (‘‘routines’’) within the pro-
gram; and installing the update only requires replacing the modified func-
tions in the program code and removing those that have become superflu-
ous (outdated). In that case, the function is the unit of substitution. Yet
another option is to include in an update package only the individual
modified lines of program code together with a list of to-be-removedlines, thus sizing the substitution units down to the level of individual
lines of code.
The three options mentioned in this example presuppose di¤erent sub-
stitution algorithms. In the first case, the complete code of the current
(original) program version is replaced by the new code. In the second
case, the update package includes a list of addresses of the to-be-replaced
and to-be-removed functions, with replacements for the former; and the
substitution algorithm performs the modifications, probably in the orderof mention of the addresses. The substitution algorithm for the third type
of substitution unit may be similar, except that the addresses now refer to
individual lines of code.
Now I can define the notions of incremental and nonincremental up-
dating. An updating process is nonincremental if the substitution units
coincide with the target domain; it is incremental if the substitution units
are smaller than the target domain, thus enabling partial updates. Incre-
mental updates usually cost less transmission and execution time, andtherefore tend to be more e‰cient than nonincremental ones.
It is not hard to see that the structural updates in self-repairs are often
incremental. Consider the self-repair in example (9a) whose target do-
main spans the complete clause before the editing term. Rather than for-
mulating and transmitting an updated version of the entire clause, the
speaker selects the major constituents of the clause as substitution unit
and includes only the two modified exemplars in the update package.
3. Structural updating in a model of the sentence production process
Structural updating may be viewed as a special mode of operation of the
human sentence production mechanism aiming at processing economy. In
order to characterize this mode of operation more precisely, I now sketch
Levelt’s (1989) widely accepted ‘‘model of the speaker’’ (see also Kempen1977, 1987; Kempen and Huijbers 1983; Kempen and Hoenkamp 1987).
According to this model, sentence production is a five-stage process: in-
tending, conceptualizing, grammatical encoding, phonological encoding,
660 G. Kempen
and articulating. In this article, I leave the fifth stage, which deals with the
phonetic aspects of sentence production, out of consideration.
The first stage of sentence production — called ‘‘thinking for speaking’’
(Slobin 1996) or simply ‘‘intending’’ — comprises the creation of a com-
municative intention. During the second stage, the ‘‘conceptualization’’
process maps the intention onto a set of concepts and conceptual (the-
matic) relations retrieved from the ‘‘mental lexicon’’. Output from thisstage is a conceptual structure (‘‘message’’) in the form of a tree with
branches whose linear order is undefined. Conceptual messages are input
to the ‘‘grammatical encoding’’ stage.
The mental lexicon also contains lemmas — ‘‘syntactic words’’ that will
be attached to syntactic trees as terminal nodes. I assume a one-to-one
correspondence between concepts and lemmas. Associated with every
lemma is information specifying how it can be used in sentences. For
instance, a lemma that corresponds to an action concept, specifies howthe concept’s thematic relations are mapped onto grammatical functions
(often: actor onto subject, patient/theme onto direct object, location onto
adverbial or prepositional modifier, etc.). Based on this information, the
grammatical encoder computes the hierarchical (dominance) and the lin-
ear (precedence) structure of the sentence. Within the grammatical en-
coder, I assume that one processing component is responsible for comput-
ing the unordered hierarchical structure, and another one for linearizing
the branches of the syntactic tree (cf. Kathol [2000] for a similar divisionof labor within an HPSG-type formalism; see also Harbusch and Kempen
2002). As part of the tree formation process, lemmas are adorned with
features whose values, during the fourth stage, guide the selection of in-
flectional properties of every terminal node and mark the position of sen-
tence accents.
Taking linearly ordered trees as input, the ‘‘phonological encoder’’ lays
the groundwork for the spoken form of sentences. In response to the in-
flectional information attached to terminal nodes, the lemmas are re-placed by lexemes (phonological wordforms). Lexemes are a third type
of lexical entries in the Mental Lexicon, in addition to concepts and lem-
mas. The phonological encoder also computes an (abstract) intonation
contour based on — among other things — the sentence accent marks
placed during the second and third stages.
Given this blueprint of the sentence production process, the similarities
between generating self-repairs and (a subset of ) coordinate structures
can be summarized as in Table 1.One of the goals of the present article is to show that the phenomena
of FCR and gapping can be understood as originating from the same
source as self-repairs of the appropriateness type: incremental structural
Clausal coordination and coordinative ellipsis 661
updating processes. I will focus on clausal coordination here, although
the approach may advance our understanding of coordination at other
hierarchical levels (NPs, PPs, APs) as well.
4. Forward conjunction reduction and gapping in the sentence production
model
If the sentence production model and the hypothesis of elliptical coordi-
nation as incremental updating are basically correct, then the following
assumption is plausible as well. Every stage of the sentence production
process should be able to work ‘‘in update mode’’: to assemble and out-
put to the next stage only novel structure, i.e., only the fragments that re-
flect the replaced or augmented conceptual content. The unchanged frag-ments can be left untouched and need not be forwarded to the next stage.
In case of gapping, the conceptual content underlying the verb is shared
between conjuncts. This also holds for the thematic relations contracted
by the verb and for the mappings between thematic relations and gram-
matical functions. Only some arguments or adjuncts need to be replaced
Table 1. Generating self-repairs and coordinate structures: a comparison
Self-repair Coordination
Intentional and conceptual updates: (part of )
the content of the reparandum is replaced by
novel content
Intentional and conceptual updates: (part
of ) the content of the reparandum is
augmented by novel content
Grammatical encoding of the revised
conceptual content begins
Grammatical encoding of the augmented
conceptual content begins
Speaker announces upcoming correction by
inserting an editing term
Speaker announces upcoming
augmentation by inserting a coordinating
conjunction
Retracing repairs: Grammatical encoding of
revised conceptual content continues.
Determination of resumption point by
comparing the reparandum with the
encoding of the revised content
FCR: Grammatical encoding of
augmented conceptual content continues.
Determination of resumption point by
comparing the first conjunct with the
encoding of the augmented content
Substitution repairs: Grammatical encoding
of revised conceptual content continues.
Gapping: Grammatical encoding of
augmented conceptual content continues.
Retracing repairs: Phonological encoding and
overt expression of updated text rightward of
the resumption point (‘‘repair text’’)
FCR: Phonological encoding and overt
expression of updated text rightward of
the resumption point (‘‘second conjunct’’)
Substitution repairs: Phonological encoding
and overt expression of substitute
constituents
Gapping: Phonological encoding and
overt expression of ‘remnants’.
662 G. Kempen
or added. This means gapping can come into existence already during the
conceptualization stage of sentence production. Only the newly constructed
conceptual arguments and/or adjuncts need to be selected as output,
together with markers telling the grammatical encoder which original
arguments/adjuncts they are supposed to augment. In example (9b), the
augmentation is restricted to the fillers of the actor argument and the di-
rection adjunct. Notice that the grammatical encoder need not initialize anew clause for the posterior conjunct; it only computes, in the context of
the anterior clause, the surface shape of the substitute major constituents
and delivers the product at the phonological encoder’s door. This means
that, in contradistinction to FCR (see below). Gapping does not involve
determining a resumption point.
(9) b. Then, the BMW crashed into the Volvo, and the Jaguar into the
Porsche
In case of clause-level FCR, the circumstances are radically di¤erent. As
the verb of the conjunct-under-construction di¤ers from that of the first
conjunct, it cannot be elided. Moreover, the new verb may license a very
di¤erent set of thematic relations and di¤erent mappings between the-
matic relations and grammatical functions. Hence, the conceptualization
stage does not qualify as a possible origin of clause-level FCR. But could
it originate in the grammatical encoding stage? As the new verb may in-
troduce grammatical functions that are absent from the first conjunct, orleave out functions that were present there, the only option is to construct
a new clause for the posterior conjunct, headed by the new verb, with all
constituents ordered from left to right. In update mode, grammatical en-
coding proceeds slightly di¤erently than in normal mode. Traversing the
anterior clause from left to right, the encoder checks the identity of each
major constituent and determines whether it could become a constituent
of the posterior clause in the same ordinal position. If so, the constituent
is not constructed from scratch but gets incorporated into the posteriorclause. The encoder returns to normal mode as soon as it runs into an an-
terior constituent that violates the incorporation criteria. From then on-
ward, all constituents of the posterior clause are assembled in the stan-
dard fashion. This procedure entails the determination of the resumption
point, which is located to the immediate left of the first constituent that
blocks incorporation. In (10c), for example, the constituents dominating
as a boy he can be incorporated into the posterior clause; the direct object
all novels by Scott cannot, however, due to its position rightward of theresumption point. In order to avoid repetition of this NP, the encoder
can opt for pronominalization instead, as in (10d). Finally, and crucially,
the output to the phonological encoder does not include the incorporated
Clausal coordination and coordinative ellipsis 663
left-peripheral string. In sum, FCR can plausibly be assumed to originate
at the level of grammatical encoding.
(10) c. As a boy he read all novels by Scott and devoured *(all novels
by Scott)
As a boy he read all novels by Scott
and devoured all novels by Scott
d. As a boy he read all novels by Scott and devoured *(them)
In the next section, I propose detailed accounts of FCR and gapping in
German and Dutch based on the incremental updating hypothesis and
on the presumed place of the two phenomena in the sentence production
model.
5. Gapping and FCR as incremental updating
5.1. Some definitions
I use the term ‘‘major constituent’’ (or ‘‘major phrase’’) of a clause in a
broad sense that includes:
– Heads, i.e., main, auxiliary or copula verbs. I assume that every ex-
emplar of each of these verb types is the head of a clause, and that
every finite or nonfinite clause has exactly one head.5 The subcate-
gorization frame of the head verb is one of the factors determining
which other constituents figure in the clause as obligatory or op-
tional major phrases. Auxiliaries will be treated as heads of clausesthat govern a complement clause headed by an infinitival or parti-
cipial head.
– Arguments and adjuncts of the head verb: subject, direct object, in-
direct object, predicate, complement (i.e., a finite or nonfinite sub-
ordinate clause), adverbial modifier, and particle.6
– Subordinating conjunctions, i.e., complementizer (in a complement
clause: that, whether) or subordinator (in an adverbial clause: while,
although, when, etc.).
As will become clear in the next subsection, we need an extended defi-
nition of major constituent to deal with a special form of gapping called
‘‘long-distance gapping’’ (LDG). The target domain of clausal coordina-
tion and ellipsis (for short: ‘‘clausal coordination domain’’) is not neces-sarily a simple clause including only one verb. Empirical observations to
be addressed below necessitate the introduction of coordination domains
spanning a hierarchy of several complement clauses. For want of a better
664 G. Kempen
term, I will refer to such a hierarchy as a superclause. A superclause is a
clause of any type (main or subordinate; finite or nonfinite; complement,
adverbial or relative) with all its major constituents except for major con-
stituents consisting of clauses that include a subordinating conjunction. For
example, sentence (11a) forms one superclause together with the stack of
two nested object complement clauses. The highest complement clause
(headed by want) is a superclause that includes the complement headedby sell. The latter is a ‘‘simple’’ superclause without any embedding (see
the strings surrounded by ‘‘[SC . . . ]’’). Variant (11b) is also a superclause
but it does not include the object complement or the adverbial clause
(both printed in italics), which open with a subordinating conjunction.
(11) a. [SC Jan expects the thief [SC to want [SC to sell the stolen
bike]]]b. [SC Jan expects [SC that the thief wants to sell the stolen
bike] [SC because . . . ] . . . ]
The above list of major constituents includes complement clauses as ma-
jor constituents of clauses headed by complement-taking verbs. The no-
tion of superclause enables a slightly broadened definition of ‘‘major con-
stituent’’: The major constituents of complement clauses that belong to the
same superclause, all count as major constituents of the superclause. For
instance, superclause (11a) includes as major constituents to want to sell
the stolen bike, to sell the stolen bike and the stolen bike. In (11b), both
the stolen bike and to sell the stolen bike qualify as major constituents in
the superclause headed by wants (but not in the superclause headed by
expects).
The left periphery of a superclause covers the string of major phrases
preceding the resumption point in FCR, i.e., the region before the firstto-be-updated major constituent of the current superclause. Non-updated
major phrases in the left periphery can be ‘‘borrowed’’ by the posterior
conjunct (e.g., the phrases containing as a boy he in example [10c]). A
borrowed major constituent functions as a normal (overt) constituent of
the borrowing (posterior) clause. In FCR, Left-peripheral major constitu-
ents are borrowed obligatorily. In gapping, borrowing of all non-updated
major constituents is obligatory.
5.2. Forward conjunction reduction and gapping
Superclauses are the application domains of incremental updating. Any
superclause can spawn a clausal coordination and become its anterior
Clausal coordination and coordinative ellipsis 665
conjunct. The shape of the posterior conjunct depends crucially on the
nature of the substitution units and the substitution algorithm. I propose
to view the major constituent as the fundamental substitution unit in
FCR and gapping. That is, whenever at least one word of a major con-
stituent of a superclause is outdated and needs to be replaced, this constit-
uent will be updated in its entirety.
In line with the assumption (Section 3) that gapping originates at anearlier stage than FCR, the update process first checks whether the set of
outdated (i.e., to-be-updated) major constituents does not contain the
head verb of the superclause (i.e., the head of the highest clause in the
stack). If so, the posterior conjunct will consist of updated versions of
all outdated major constituents, possibly extended with additional major
phrases that have no counterpart in the anterior conjunct (e.g., Mary
sang, and beautifully!). All non-outdated major phrases are obligatorily
borrowed by the posterior conjunct (i.e., may not reappear there). Hence,the posterior conjunct consists exclusively of a string of remnants.
If the ‘‘highest’’ head verb of the superclause does need an update, then
all left-peripheral major constituents are borrowed obligatorily (and no
constituents rightward of the resumption point). The result is FCR.
To prevent a potential misunderstanding, I should point out that
speakers are free to choose between incremental or non-incremental up-
dates of the current superclause. For instance, they can repeat the two ini-
tial words of (12a) in the posterior conjunct or leave them out. However,if the update proceeds in incremental mode, the substitution algorithm
dictates which major constituents are allowed to show up in the posterior
conjunct. In (12a)–(12b), the temporal modifier vanmorgen is not allowed
to become a remnant because it is not outdated.7
(12) a. *omdat Jan vanmorgen een DVD leende en
because Jan this-morning a DVD borrowed and
omdat Jan vanmorgen een CD kocht
because Jan this-morning a CD bought
b. *Jan leende vanmorgen een DVD van Marie
Jan borrowed this-morning a DVD from Marie
en vanmorgen een DVD van Anna
and this-morning a DVD from Anna
In the remainder of this Section, I apply incremental updating to the
key elliptical coordination phenomena in simple clauses (i.e., super-clauses consisting of a single clause without subordinate clauses) and in
nested clauses spanning one or more superclauses. The target language is
Dutch.
666 G. Kempen
5.2.1. Incremental updating in simple clauses. The examples in (13)
each contain two coordinated clauses. While no ellipsis has been applied
to (13a), the other versions illustrate FCR and gapping, respectively
(13) a. Jan koopt auto’s en Peter steelt fietsen
‘Jan buys cars and Peter steals bikes’
Jan koopt auto’sen Peter steelt fietsen
b. Jan koopt auto’s en steelt fietsen (FCR)
‘Jan buys cars and steals bikes’
Jan koopt auto’s
en steelt fietsen
c. Jan koopt auto’s en Peter fietsen (gapping)
‘Jan buys cars and Peter, bikes’
Jan koopt auto’sen Peter ......... fietsen
The sentences in (14) illustrate that borrowing in FCR is left peripheral
only, as predicted by incremental updating. In (14a), the posterior con-
junct borrows its subject NP Jan from left. Variant (14b) is ungrammatical
because the direct object fietsen cannot be borrowed right-peripherally,
implying the obligatory presence of an overt direct object (here: the pro-
noun ze ‘them’). In subordinate clause (14c), with SOV word order, theposterior clause can legally borrow the direct object, which now is within
the left periphery. The examples in (15) confirm the constraint on left-
peripheral borrowing in FCR.
(14) a. Jan koopt fietsen en steelt ze
‘Jan buys bikes and steals them’
Jan koopt fietsen
en steelt zeb. *Jan koopt fietsen en steelt ze
* Jan koopt fietsen
en steelt *******
c. dat Jan fietsen koopt en ze steelt
that Jan bikes buys and them steals
‘that Jan buys bikes and steals them’
dat Jan fietsen koopt
en ze steelt(15) a. *dat Peter fietsen steelt en Jan fietsen koopt
* dat Peter fietsen steelt
en Jan ***** koopt
Clausal coordination and coordinative ellipsis 667
b. *Dit soort fietsen steelt Peter en verkoopt hij
this kind bikes steals Peter and sells he
‘This kind of bikes Peter steals and sells’*Dit soort fietsen steelt Peter
en verkoopt ****
I conclude this subsection with some remarks on gapping. The sentences
in (16) show that gapping allows nonidentical inflectional forms of thesame word to be elided: The di¤erence between schrijven and schrijft is ir-
relevant. This is in line with an essential assumption underlying the gap-
ping mode of incremental updating, viz. that the updating process only
involves creating and delivering substitutes for the inappropriate (contras-
tive) major constituents and leaves the other ones untouched.
(16) a. Wij schrijven artikelen en hij schrijft een boek
we write articles and he writes a book
‘We write articles and he writes a book’
Wij schrijven artikelen
en hij .............. een boek
b. *Wij schrijven artikelen en hij schrijft een boek
* Wij ****** artikelen
en hij schrijft een boekc. . . . dat wij artikelen schrijven en hij een boek schrijft
dat wij artikelen schrijven
en hij een boek .............
Furthermore, notice that gapping-style borrowing also applies to ‘‘Argu-
ment cluster coordination’’ (cf. Steedman 2000: Ch. 7; Beaver and Sag2004), where the to-be-updated constituents of the anterior clause are ad-
jacent, as in (16c) and (17), rather than separated by the head verb. The
incremental updating process is indi¤erent to the position of the head
verb in the anterior conjunct. Hence, argument cluster coordination does
not need a special treatment.
(17) We geven Susan een mobieltje en Patrick een iPod
‘We give Susan a cell phone and Patrick an iPod’
we geven Susan een mobieltje
en ............. Patrick een iPod
Finally, the linear order of gapping remnants need not be identical to that
of their counterparts in the anterior conjunct. Examples like (18) show
that both orders of the remnants morgen en iemand uit de buurt yield
good sentences, irrespective of the order of their counterparts ik and van-
daag. Versions (18a) and (18c) with the subject NP iemand uit de buurt in
668 G. Kempen
final position within the posterior conjunct, sound even slightly better
than versions (18b) and (18d) with the subject in initial position. Presum-
ably, factors relating to information structure or constituent length, or
both, are responsible for this e¤ect. The indefiniteness of iemand uit de
buurt may create a preference for a late position — a tendency reinforced
here by ‘‘heavy NP shift’’ due to length.
(18) a. Ik let vandaag op de poes, en morgen
I look today after the cat and tomorrow
iemand uit de buurt
someone from the neighborhood
‘I look after the cat today, and tomorrow someone from the
neighborhood’
b. Ik let vandaag op de poes, en iemand uit de buurt morgen
c. Vandaag let ik op de poes, en morgen iemand uit de buurt
d. Vandaag let ik op de poes, en iemand uit de buurt morgen
The phenomenon of remnant order reversals in gapping can be under-
stood in terms of the sentence production model put forward in Sections
3 and 4. According to this model, the dominance relations between the
nodes of a syntactic tree are computed by a di¤erent component of thegrammatical encoder than the linear order of the branches of the tree.
After the grammatical encoder has received as input the updated concep-
tual content for the outdated major constituents, the hierarchical compo-
nent can compute the shape of the remnants such that they fit into the hi-
erarchical structure of the anterior clause. The linear order component
then determines the precedence relations among the branches of the
modified tree in accordance with information-structural and other con-
straints. At this point, modified information-structural conditions maygive rise to precedence relations in the posterior conjunct that di¤er from
those in the anterior clause. As its last step, the grammatical encoder does
not pass the complete tree on to the phonological encoder but only the
updated constituents (remnants) in their then left-to-right order.
5.2.2. Incremental updating in nested clauses. The definition of super-
clause in Section 4.1 implies that adverbial subordinate clauses, whether
finite or nonfinite, launch their own superclause and constitute their own
incremental updating domain. This is due to the obligatory subordinating
conjunction at the onset of such a clause. Hence, gapping is ruled out ifone of the remnants pairs up with a constituent belonging to the main
clause, and the other one to an embedded adverbial clause. Consequently,
examples (19b) and (20b) are ill-formed, in contrast with variants (19a)
Clausal coordination and coordinative ellipsis 669
and (20a), where the remnants are complete major constituents of the
main clause. (The brackets ‘‘[SC . . . ]’’ indicate superclause boundaries.)
(19) a. Jan vertrok toen zijn ex aankwam en Peter toen zijn
‘Jan left when his ex arrived and Peter when hiscollega aankwam
colleague arrived’
Jan vertrok [SC toen zijn ex aankwam]
en Peter .......... [SC toen zijn collega aankwam]
b. *Jan vertrok toen zijn ex aankwam en Peter zijn collega
* Jan vertrok [SC toen zijn ex aankwam]
en Peter ......... [SC **** zijn collega *********]
(20) a. Jan vertrok om zijn vrouw op te halen en
Jan left in-order his wife up to pick and
Peter om zijn collega op te halen
Peter in-order his colleague up to pick
Jan vertrok [SC om zijn vrouw op te halen]
en Peter ........... [SC om zijn collega op te halen]
b. *Jan vertrok om zijn vrouw op te halen en Peter zijn collega
* Jan vertrok [SC om zijn vrouw op te halen]
en Peter .......... [SC *** zijn collega **********]
Neijt (1979: 183) gives several examples showing that the same pattern
holds for complement clauses introduced by a subordinating conjunction
(complementizer) — see (21) and (22a). The nonfinite complement in
(22b), however, without a complementizer, does not launch its own super-clause, thus licensing long-distance gapping.
(21) *Jan veronderstelt dat hij een novelle zal schrijven en
Jan assumes that he a novel will write and
Peter veronderstelt dat hij een toneelstuk zal schrijven
Peter assumes that he a theater-play will write*‘Jan assumes that he will write a novel and Peter that he will
write a theater play’
* Jan veronderstelt [SC dat hij een novelle zal schrijven]
en Peter .................. [SC ****** een toneelstuk ************]
(22) a. *Kees probeerde om Bernhard te imiteren en
Kees tried CMPR Bernhard to imitate and
Harry probeerde om Fred te imiteren
Harry tried CMPR Fred to imitate‘Kees tried to imitate Bernhard and Harry, Fred’
* Kees probeerde [SC om Bernhard te imiteren]
en Harry .............. [SC *** Fred **********]
670 G. Kempen
b. Kees probeerde Bernhard te imiteren en Harry probeerde Fred
te imiteren
‘Kees tried to imitate Bernhard and Harry, Fred’[SC Kees probeerde Bernhard te imiteren
en Harry ............... Fred ................]
The coordinations in (23) all contain the head verb vragen ‘ask’ governing
an extraposed infinitival complement without complementizer, and in
each of them the anterior conjunct qualifies as a superclause. While sen-
tence (23a) illustrates long-distance gapping, variants (23b) and (23c) are
‘‘simple’’ gapping cases since their complements have been updated intheir entirety (not only their head verbs but also their direct objects — a
type of gapping called subgapping. In (23d), the verb of the complement
clause has been updated but not the direct object de tekst. That is to say,
the update does cover not the entire complement, and gapping fails. FCR
is ruled out as well because the direct object is not located in the left pe-
riphery. Versions (23e) and (23f ), both without direct object updates, can
be analyzed as legal FCR structures since they do allow left-peripheral
borrowing of the direct object — with the complement clause or themain clause as coordination domain in (23e) and (23f ), respectively.
(23) a. Hij vraagt mij de tekst te controleren en jou
he asks me the text to check and you
de literatuurlijst
the references
‘He asks me to check the text and you, the references’
Hij vraagt mij de tekst te controleren.............. en jou de literatuurlijst ....................
b. Hij vraagt mij de tekst te controleren en jou
he asks me the text to check and you
de literatuurlijst te actualiseren
the references to update
‘He asks me to check the text and you, to update the
references’
Hij vraagt mij de tekst te controleren.............. en jou de literatuurlijst te actualiseren
c. . . . dat hij mij vraagt de tekst te controleren en
. . . that he me asks the text to check and
jou de literatuurlijst te actualiseren
you the references to update
‘. . . that he asks me to check the text and you, to update the
references’
Clausal coordination and coordinative ellipsis 671
dat hij mij vraagt de tekst te controleren
......... en jou ......... de literatuurlijst te actualiseren
d. *Hij vraagt mij de tekst te controleren en jou
he asks me the text to check and you
de tekst te verspreiden
the text to distribute
*‘He asks me to check the text and you, to distribute the text’
*He vraagt mij de tekst te controleren
.............. en jou ****** te verspreiden
e. Hij vraagt mij de tekst te controleren en de
he asks me the text to check and thetekst (te) verspreiden
text to distribute
‘He asks me to check the text and to distribute it’
Hij vraagt mij [SC de tekst te controleren
en te verspreiden]
f. Deze tekst vraagt hij mij te controleren en (te)
this text asks he me to check and to
verspreiden
distribute
‘This text, he asks me to check and (to) distribute’
[SC Deze tekst vraagt hij mij te controleren
en te verspreiden]
The sentences in (24a)–(24f ) run parallel to (23a)–(23f ) except that their
main clauses are headed by a modal verb instead of an extraposition verb.
They reveal that clause union does not modify the pattern of grammati-
cality ratings.
(24) a. Ik moet de tekst controleren en jij de
I should the text check and you the
literatuurlijst
references
‘I should check the text and you, the references’
Ik moet de tekst controleren
en jij ........ de literatuurlijst .................b. Ik moet de tekst controleren en jij de
I should the text check and you the
literatuurlijst actualiseren
references update
‘I should check the text and you should update the references’
Ik moet de tekst controleren
en jij ....... de literatuurlijst actualiseren
672 G. Kempen
c. . . . dat ik de tekst moet controleren en jij
. . . that I should the text check and you
de literatuurlijst actualiseren
the references update
‘. . . that I should check the text and you should update the
references’
dat ik de tekst moet controleren
.... en jij de literatuurlijst ....... actualiseren
d. *Ik moet de tekst controleren en jij de tekst
I should the text check and you the text
verspreiden
distribute
‘I should check the text and you should distribute it’
* Ik moet de tekst controleren
en jij .........****** verspreiden
e. Ik moet de tekst controleren en de tekst
I should the text check and the text
verspreiden
distribute*‘I should check the text and distribute the text’
Ik moet [SC de tekst controleren
en verspreiden]
f. Deze tekst moet ik controleren en verspreiden
this text should I check and distribute
‘This text, I should check and distribute’
[SC Deze tekst moet ik controleren
en verspreiden]
The definition of superclause allows long-distance gapping to cut into
dependent Wh-questions that do not contain a minor constituent — see
(25a). Variant (25b) shows that the Wh-phrase is an obligatory member ofthe posterior conjunct. I assume that this restriction need not be stipulated
explicitly but follows from the general prohibition against elision
of contrastive expressions in gapping: Although the two referring expres-
sions (here: Wh-phrases) are identical, non-coreferentiality is presupposed.
Version (25c) shows that of ‘whether’ does not count as a Wh-phrase but as
a subordinating conjunction and hence introduces its own superclause.
(25) a. sommigen vragen wie Bach speelt en anderen
some ask who Bach plays and others
vragen wie Buxtehude speelt
ask who Buxtehude plays
‘Some ask who plays Bach and others who (plays) Buxtehude’
Clausal coordination and coordinative ellipsis 673
[SC Sommigen vragen wie Bach speelt
en anderen ......... wie Buxtehude ........]
b. *Sommigen vragen wie Bach speelt en anderen vragen wie
Buxtehude speelt
c. *Sommigen vragen of we Bach spelen en anderen
some ask if we Bach play and others
vragen of we Buxtehude spelen
ask if we Buxtehude play
*‘Some ask whether we can play Bach and others Buxtehude’
* Sommigen vragen [SC of we Bach spelen]
en anderen ......... [SC ***** Buxtehude ****]
A theoretical issue posed by the superclause as incremental updating do-
main arises from Wh-fronting. A Wh-phrase may escape from the com-
plement clause it belongs to, and land at the beginning of a clause higherup in the hierarchy. Do such extracted constituents count as members of
the superclause they originate from, or of the one where they have land-
ed? Example (26a) is a case in point: The interrogative prepositional
phrase op welke dagen ‘on which days’ can be plausibly interpreted only
as a temporal modifier of open zijn ‘be open’ — it must have been ex-
tracted from the finite complement clause introduced by that. The sen-
tence is well-formed although, of the contrastive constituent pairs, one is
physically located in the main clause (the Wh-phrases) and the other onein the complement (schoenenzaak ‘shoe store’ and modezaak ‘fashion
store’). Now compare (26c), which has contrastive pairs at the same linear
positions but where the interrogative PP has not been extracted from the
finite complement (tegen wie ‘to whom’ is only interpretable as an argu-
ment of gezegd ‘said’). The ungrammaticality of this variant is predicted
on the theory that superclauses are the domain of incremental updating:
The main clause and the complement are di¤erent superclauses, which
rules out long-distance gapping. This, in turn, entails well-formedness of(26d), where the contrastive pairs belong to the same superclause. (Nei-
ther the pair of subject NPs u–uw partner nor the pair of Wh-phrases
have been extracted.) In line with this reasoning, version (26b) is ungram-
matical as the two contrastive pairs belong to the di¤erent superclauses.
(This applies to the intended — and the only meaningful — interpreta-
tion, with uw partner ‘your partner’ fulfilling subject role in the main
rather than the embedded clause.)8
(26) a. Op welke dagen heeft u gezegd dat de
On which days have you said that the
schoenenzaak open moet zijn, en op welke dagen
shoe-store open should be, and on which days
674 G. Kempen
heeft u gezegd dat de modezaak open moet
have you said that the fashion-store open should
zijn?
be
‘On which days did you say that the shoe store should be
open, and on which days did you say that the fashion store
should be open?’
b. *Op welke dagen heeft u gezegd dat de
on which days have you said that the
schoenenzaak open moet zijn, en op welke dagen
shoe-store open should be, and on which daysheeft uw partner gezegd dat de schoenenzaak
has your partner said that the shoe-store
open moet zijn?
open should be
‘On which days did you say that the shoe store should be
open, and on which days did your partner say that the shoe
store should be open?’
c. *Tegen wie heeft u gezegd dat de
to whom have you said that the
schoenenzaak vaker open moet zijn, en tegen
shoe-store more-often open should be, and to
wie heeft u gezegd dat de modezaak
whom have you said that the fashion-store
vaker open moet zijn?
more-often open should be
‘To whom did you say that the shoe store should open moreoften, and to whom did you say that the fashion store should
open more often?’
d. Tegen wie heeft u gezegd dat de
to whom have you said that the
schoenenzaak vaker open moet zijn, en tegen
shoe-store more-often open should be, and to
wie heeft uw partner gezegd dat de
whom has your partner said that theschoenenzaak vaker open moet zijn?
shoe-store more-often open should be
‘To whom did you say that the shoe store should open more
often, and to whom did your partner say that the fashion
store should open more often?’
Clausal coordination and coordinative ellipsis 675
In conclusion, Wh-fronting — and, presumably, any form of A-bar
movement — does not a¤ect superclause membership of the moved
phrases. This confirms the validity of the definition of superclause in
terms of the hierarchical rather than the linear-order relations between
clauses and their constituents.
6. Backward conjunction reduction
At first sight, BCR looks like the mirror image of FCR. On closer inspec-
tion, however, this impression is false — for several reasons. First, BCR
requires an accented word or word group immediately adjacent to the
elided fragment. This holds even in the absence of semantic or pragmatic
reasons for accentuation, as in (27a) where the prepositions op and naar
carry an obligatory contrastive accent although their meanings are not
contrastive at all. In FCR variant (27b), no contrastive accent is needed
on luistert — the word immediately adjacent to the elided string. (See
also [30a] below.)
(27) a. Mijn kamergenoot is dol op klassieke muziek,
my roommate is fond of classical music
en luistert vaak naar klassieke muziek
and listens often to classical music
‘My roommate is fond of, and often listens to, classical music’
b. Mijn kamergenoot is dol op klassieke muziek en
my roommate is fond of classical music andmijn kamergenoot luistert er vaak naar
my roommate listens there often to
‘My roommate is fond of classical music and often listens to
it’
Second, FCR requires lemma rather than lexeme identity of the elided
string. For instance, consider the German BCR example in (28a), with
the verbs sagen ‘say’ and fragen ‘ask’ in the anterior and posterior con-
juncts, respectively. Both verbs are ditransitive, but their indirect objects
take di¤erent case: dative (sagst mir) versus accusative ( fragst mich). The
di¤erence between the indirect object lexemes (mir/mich) does not rule
out BCR: elision of mir nie etwas in the anterior conjunct has, at worst,
a marginal e¤ect on grammaticality. In synonymous FCR variant (28b),however, with fronting of the indirect objects, the lexeme di¤erence has
fatal consequences, despite the fact that mir is in the left-periphery of the
anterior clause and borrowable.
676 G. Kempen
(28) a. Du sagst mir nie etwas und du fragst mich
you tell me never anything and you ask me
nie etwas
never anything
‘You never tell or ask me anything’
b. *Mir sagst du nie etwas und mich fragst du nie etwas
‘Me, you never tell or ask anything’
Third, BCR allows elision of incomplete major constituents whilst FCR
does not. In the anterior conjunct of Dutch FCR example (29a), the final
word of the subject NP can be elided without loss of grammaticality; its
mirror image BCR counterpart (29b), however, becomes ill-formed if the
first word of the subject NP is suppressed in the posterior conjunct.
(29) a. Nu arriveren twee treinen en vertrekken drie
Now arrive two trains and leave three
treinen
trains
‘Now, two trains arrive and three trains leave’
b. *Twee treinen arriveren nu en twee bussen vertrekken nu
‘Two trains arrive now and two busses leave now’
Fourth, BCR does not require the elided fragment to fulfill the same
grammatical function as its counterpart in the posterior conjunct, where-
as FCR does. The elided string de zoon van mijn buurman of (30a)’s ante-rior conjunct plays the role of subject NP but its namesake in the poste-
rior clause is the complement of a preposition. This di¤erence does not
degrade the sentence. In example (30b), FCR is ruled out due to the dif-
ference between the grammatical functions of de zoon van mijn buurman:
subject in the anterior, direct object in the posterior clause. Example (30a)
also illustrates that, in BCR, the elided string should be preceded immedi-
ately by a constituent that carries contrastive stress — even if this stress
does not have a semantic origin: In a nonelliptical version, neither dan
nor van would have been accented.
(30) a. Ik ben kleiner dan de zoon van mijn buurman
I am taller than the son of my neighbor
en verlies dus gewoonlijk van de zoon van
and lose therefore usually to the son of
mijn buurman
my neighbor
‘I am smaller than, and therefore usually lose to the son of my
neighbor’
Clausal coordination and coordinative ellipsis 677
b. *De zoon van mijn buurman is groter dan ik
the son of my neighbor is taller than I
en de zoon van mijn buurman versla ik dus
and the son of my neighbor beat I therefore
zelden
seldom
‘My brother is taller than I and therefore I seldom beat him’
Fifth, the string that is borrowed by the anterior conjunct, need not be
(right-)peripheral in the posterior conjunct. In the posterior clause of
(31), the overt token of mijn optreden is followed by the string weer terug
te zijn, which cannot be borrowed by the anterior conjunct because the
clause Ik vertrek twee dagen voor mijn optreden weer terug te zijn is ill-
formed. In (32), the overt token of het Europese gemiddelde is not periph-
eral due to the clause-final particle uit. Remember that in FCR the bor-
rowed string has to be (left-)peripheral in the anterior clause.
(31) Ik vertrek twee dagen voor mijn optreden en
I leave two days before my performance and
hoop drie dagen na mijn optreden weer terug
hope three days after my performance again back
te zijn
to be
‘I leave two days before, and hope to be back again three days
after my performance’
(32) De score van Nederland begint op een niveau ver
the score of the-Netherlands begins at a level far
boven het Europese gemiddelde, en komt uiteindelijk
above the European average and ends eventually
net onder het Europese gemiddelde uit.
just below the European average up
‘The score of The Netherlands begins at a level far above, and
eventually ends up just below the European average’
Incidentally, this obervation has an important theoretical consequence.
Since Postal (1974), many accounts of BCR have been based on ‘‘rightnode raising’’ or an equivalent mechanism. However, such accounts pre-
suppose that the overt counterpart of the elided string is right-peripheral
in the posterior conjunct. Hence, they are incompatible with the well-
formedness of sentences such as (31) and (32).
All five BCR properties highlighted here are shared by an elliptical
device that is also operative outside of coordinate structures (Hudson,
1976). Following Kathol (1999), I call this mechanism Left Deletion
678 G. Kempen
(LD). As illustrated by (33b), from Haeseryn et al. (1997: 1562), LD may
elide right-peripheral material shared by a subordinate clause and the
main clause following it, or even material shared the subject and the di-
rect object of the same clause (34).
(33) a. Hij is vol hoop op een goede uitslag, maar
he is full-of hope for a good result, but
niet afhankelijk van een goede uitslag
not dependent on a good result
‘He is full of hope for, but not dependent on a good result’
b. Hoewel hij vol hoop is op een goede uitslag, is hij niet
afhankelijk van een goede uitslag
‘Although he is full of hope for a good result, he is not
dependent on a good result’
(34) a. Een man met kinderen, huwt graag een vrouw
a man wıth children, marries gladly a woman
zonder kinderen
without children
‘A man wıth, likes to marry a woman without children’b. . . . dat een man met kinderen, graag een vrouw zonder kinderen
huwt
‘. . . that a man wıth, likes to marry a woman without children’
These five points of similarity strongly suggest that BCR and LD shouldbe treated as one and the same phenomenon. If this hypothesis is correct
— which I assume — it follows that BCR is not a form of coordinate
ellipsis: Coordinate structures only a¤ord a suitable playing ground for
Left Deletion as they often give rise to contrastive pairs. The plausibility
of viewing BCR as a form of coordinate ellipsis, hence of incremental up-
dating, is extremely low anyway: the notion of updating entails forward
ellipsis only because, by definition, the update comes later than the origi-
nal structure.An interesting question is raised by the directionality of BCR/LD:
Why does it elide backward rather than forward? I suggest two possible
— both performance-related — reasons. First, consider that the elided
string and its overt counterpart are always immediately preceded by a
contrastively accented word or word group (e.g., voor ‘before’ and na
‘after’ in (31]). Eliding the first token of the string brings the two con-
trasted words or word groups closer together, thus enhancing the con-
trast. This perceptual e¤ect probably benefits the listener/reader. Theother reason has to do with the grammatical encoding process and may
benefit the speaker. BCR helps to avoid reduplication of stretches of
speech within the same sentence. One strategy to avoid reduplications is
Clausal coordination and coordinative ellipsis 679
BCR, another one is based on pronominalization. As pronouns usually
follow their antecedents, the latter strategy often reaches its goal by by
selecting a concise pronoun. This virtually eliminates the need for a re-
versed BCR/LD.
How does BCR/LD fit into the model of the speaker? The fact that
only right-peripheral subconstituents of the anterior conjunct/constituent
can be elided, entails that BCR/LD can originate no earlier than duringthe grammatical encoding stage of sentence production — more precisely:
after linear order of the subconstituents has been established. Prosodic
properties, in fact, imply that the origin of BCR/LD must be even later,
namely in the phonological encoding stage. Remember from examples
such as (27) and (30) that the nonelided remnant of the anterior
conjunct/constituent must end in a contrastively accented lexical item,
even if this accent is not rooted in the meaning of the utterance. The latter
condition implies that the accent cannot have been prepared during anearlier stage. The conclusion must be that BCR/LD originates in the
course of computing the sentence’s intonation contour by the phonologi-
cal encoding component.
This ‘‘late accenting’’ phenomenon also provides a counterargument
against an otherwise appealing theoretical possibility of relating BCR to
self-repairs. In Kempen (1991), I proposed to view BCR as the coordina-
tion analog of the interrupt stage of self-repairs, that is, of the stage where
the speaker withholds the remainder of a planned utterance in response tospotting an error in the realized part of the utterance. It is obvious, how-
ever, that such interrupts often occur before the speaker has planned the
repair text. In contrast, speakers who produce a BCR or an LD structure
must have carried out a considerable amount of advance planning at the
moment of interrupt: They must have (1) grammatically encoded the pos-
terior part, (2) noticed the structural similarities with the part, and (3)
inserted the required sentence accents. This di¤erence rules out a direct
theoretical correspondence between the interrupt stage of self-repairs onthe one hand, and BCR/LD on the other.
Before leaving the topic of BCR, I should point out a remarkable inter-
action between BCR and gapping. Sentence (35a) is a candidate for BCR
(due to the right-peripheral hun tentamens being shared by the conjuncts)
as well as gapping (due to the shared verb doen). Putting both options into
e¤ect yields (35b). Variant (35c) features gapping only. As mentioned in
connection with example (18) at the end of Section 4.2.1, the order of the
gapping remnants need not correspond exactly to that of their counterpartsin the anterior conjunct. Hence, sentence (35d), with the order of de twee-
dejaars and eind juni reversed, is predicted to be well-formed — in agree-
ment with fact. In combination with BCR, this word order freedom is dras-
680 G. Kempen
tically reduced due to the requirement that the elided string in the anterior
conjunct and its overt counterpart in the posterior conjunct have to be pre-
ceded by accented constituents belonging to the same contrast pair. This
condition is violated in (35e) where eind mei does not pair up with de twee-
dejaars. The explanation of this phenomenon awaits further investigation.
(35) a. De eerstejaars doen eind mei hun tentamens
the 1st-yr-students take end May their exams
en de tweedejaars doen eind juni hun
and the 2nd-yr-students take end June their
tentamens
exams
‘The first-year students take their exams end of May and the
second-year students take their exams end of June’
b. De eerstejaars doen eind mei hun tentamens, en de tweedejaars
doen eind juni hun tentamens
c. De eerstejaars doen eind mei hun tentamens en de tweedejaars
doen eind juni hun tentamens
d. De eerstejaars doen eind mei hun tentamens en eind juni doen
de tweedejaars hun tentamens
e. *De eerstejaars doen eind mei hun tentamens, en eind juni doen
de tweedejaars hun tentamens
7. Asymmetrical coordinations in Dutch and German
The three groups of elliptical coordinations discussed above (gapping,
FCR and BCR) are often called ‘‘symmetrical’’ or ‘‘even’’ because the
elided string in one conjunct occupies the same linear position as their
overt counterpart in the other conjunct. The fourth group, SGF, which
does not meet this criterion, is therefore called ‘‘asymmetrical.’’ In thepresent section, I focus on SGF and other phenomena of coordinate ellip-
sis that exhibit asymmetry.
7.1. SGF coordination
As mentioned at the outset of the article, an important class of coordinate
constructions goes by the name of SGF (‘‘Subject gap in finite clauseswith fronted verb’’; Hohle 1983). The update diagram accompanying ex-
ample (36a) indicates that the finite verb verkocht ‘sold’ of the posterior
conjunct cannot borrow its subject NP Jan which, due to subject-verb
Clausal coordination and coordinative ellipsis 681
inversion, does not occupy a position leftward of the resumption point
and therefore cannot be borrowed. Nevertheless, the sentence is fully ac-
ceptable, with Jan interpreted as the subject of verkocht. This paradox
does not arise in FCR variant (36b) without inversion. The observation
that SGF coordination is incompatible with incremental updating sug-
gests that attempts to reduce this phenomenon to symmetrical (even) co-
ordinate ellipsis (e.g., Johnson 2002) are likely to fail. This raises thequestion how to account for SGF di¤erently.
(36) a. Gister stal Jan een fiets en verkocht hem
Yesterday stole Jan a bike and sold itvanmorgen
this-morning
‘Yesterday Jan stole a bike and sold it this morning’
*Gister stal Jan een fiets
en verkocht **** hem vanmorgen
b. Jan stal gister een fiets en verkocht hem vanmorgen
Jan stal gister een fiets
en verkocht hem vanmorgen
Various authors have noted that the conjuncts of an SGF coordination
contract a special semantic relationship (Heycock and Kroch 1994;
Sturm 1995; Frank 2002; Hendriks 2004; Reich 2008): The predicates
that these conjuncts ascribe to the referent of the subject NP, do not referto several independent events or situations but to aspects of a single, com-
pound event/situation. For instance, SGF example (37a) expresses that
visiting the bar and getting drunk necessarily co-occur. However, version
(37b), with an overt subject NP in the second conjunct, does not entail
co-occurrence: During a depressive fit, the protagonist might usually get
drunk at home, for instance.
(37) a. Tijdens depressieve buien gaat hij naar zijn
during depressive fits goes he to his
stamkroeg en bedrinkt zich
favorite-bar and gets-drunk himself
‘During depressive fits he goes to his favorite bar and getsdrunk’
b. Tijdens depressieve buien gaat hij naar zijn stamkroeg en
bedrinkt hij zich
‘During depressive fits he goes to his favorite bar and he gets
drunk’
Another characteristic semantic property of SGF examples concerns their
information structure (or ‘‘information packaging’’; cf. Vallduvı and Eng-
682 G. Kempen
dahl 1996). Hendriks (2004) observes that in felicitous SGF coordinations
the subject refers to topical (old, presupposed) information. Indeed, SGF
coordination is bad if the subject NP introduces new, foregrounded infor-
mation, as in (38). Grammaticality is fully restored by insertion of an
overt subject (e.g., hij ‘it’ in front of duurt ‘lasts’). The two semantic prop-
erties of felicitous SGF examples can be combined into the following
statement: The predicates of an SGF coordination ascribe distinct aspectsof one compound event/situation as new information to the referent of
the same subject. By implication, the subject referent cannot be new/
foregrounded in the anterior clause.
(38) Vanavond draait ‘‘L’enfant sauvage’’ van Tru¤aut
Tonight is-showing ‘‘L’enfant sauvage’’ by Tru¤aut
en *(hij) duurt ongeveer twee uur
and it lasts about two hours
‘Tonight ‘‘L’enfant sauvage’’ by Tru¤aut is showing and (it) lasts
about two hours’
These observations suggest that posterior clauses of SGF coordinations do
not borrow a subject NP in the course of an incremental updating opera-
tion. Instead, I propose that, in the communicative intention underlying
an SGF coordination, the speaker assigns several predicates to the refer-
ent of the subject NP simultaneously. In sentence (36), two actions are at-
tributed to Jan: having stolen a bike yesterday, and having sold it thismorning. In the hierarchical syntactic structure of (36), the verb verkocht
‘sold’ does not have an NP as subject but only a ‘‘coreference tag’’ point-
ing to the person called Jan. The verb stal ‘stole’ does have an explicit
subject NP, and it carries the same coreference tag. In the linear structure
of (36), the subject position of the posterior clause is empty because no
subject NP needs to be linearized in this clause.
Example (39), from Sturm (1995), indeed shows that the subject of a
posterior SGF clause cannot have had an original linear position corre-sponding to that of its counterpart in the anterior clause. Version (39a) is
an interrogative main clause whose subject je ‘you’ occupies a position
after the finite verb due to subject-verb inversion (obligatory in interroga-
tive main clauses). Analyzing this sentence as (39b) is blocked by the pres-
ence of the second-person su‰x -t on the finite verb kijkt. A general mor-
phosyntactic rule prohibits this su‰x if subject NP je follows the verb.
Hypothesizing an underlying structure without inversion, as (37c), is
ruled out as well because the second conjunct would no longer be an in-terrogative clause. Actually, (39d) is the only correct interrogative variant
with an overt subject NP in the posterior conjunct (see Heycock and
Kroch (1994) for a similar reasoning). The only way out seems to be the
Clausal coordination and coordinative ellipsis 683
assumption that the linear structure of the posterior clause does not host
a subject at all, and that in its hierarchical structure the subject role is
played by a tag pointing to the referent of the subject of the anterior
clause.
(39) a. Waarom zit je daar en kijkt zo droevig om
Why sit you there and look so sad around
je heen?
you --
‘Why are you sitting there and looking around so sad?’b. *Waarom zit je daar en kijkt je zo droevig om je heen?
c. *Waarom zit je daar en je kijkt zo droevig om je heen?
d. Waarom zit je daar en kijk je zo droevig om je heen?
Frank (2002) provides a compelling argument in support of this assump-
tion, i.e., that the subject of a posterior SGF clause merely consists of a
pointer to the subject of the anterior clause. Consider sentence (40a) —
Frank’s translation of her German SGF example (17b), reproduced here
as (40b). The meaning of this sentence implies that virtually nobody who
has bought a car will travel by bus anymore. It cannot be interpreted as
(40c) — the translation of Frank’s example (17c), reproduced here as(40d) — where the posterior conjunct hosts its own subject, implying
that virtually nobody travels by car or bus. In (40a), the set of car buyers
may be large; in (40c), it is nearly empty. In other words, the reference of
the overt almost no one in the anterior clause of (40a) is determined on the
basis of both predicates, whereas subject reference in the anterior clause
of (40c) is selected on the basis of the anterior predicate only. The well-
formedness of German (40b) shows that SGF constructions behave as
(40a), not as (40c). That die wenigsten Leute in (40b) functions both asexplicit anterior subject and as implicit posterior subject, and that these
subjects are coreferential, originates from the underlying communica-
tive intention (one subject referent, multiple predicates), not from some
form of non-left-peripheral borrowing during the grammatical encoding
process.9
(40) a. Therefore almost no one buys a car and takes the bus
b. Daher kaufen die wenigsten Leute ein Auto und
Therefore buy the fewest people a car and
fahren mit dem Bus
ride with the bus
c. Therefore almost no one buys a car and almost no one takes the
bus
d. Daher kaufen die wenigsten Leute ein Auto und fahren die
wenigsten Leute mit dem Bus
684 G. Kempen
Finally, what can we conclude with respect to the place of SGF in the
model of the speaker? The analysis put forward in this section implies,
strictly speaking, that SGF is not an elliptical phenomenon at all: At no
point in time during the production process was there a Subject NP, or a
conceptual fragment underlying a Subject NP, that got elided. Therefore,
I propose that SGF comes into existence already at the prelinguistic stage
of planning the communicative intention: attributing two predicates toone referent.
7.2. Other putatively asymmetrical coordinations in German
Recently, various purportedly asymmetrical German constructions
have received a great deal of attention (Heycock and Kroch 1994; Buring
and Hartmann 1998; Schwarz 1998; Kathol 1999; Johnson 2002). Infact, the SGF coordination discussed in the preceding subsection, is
the primus inter pares. Here, I analyze the coordinations in (41) and (42)
below, all due to Schwarz (1998) and extensively discussed by Johnson
(2002).
Example (41a) is considered asymmetrical if analyzed as (41b): Direct
object NP Die Suppe has been extracted from the anterior member of
the coordinated nonfinite clauses (VPs) but not from the posterior mem-
ber. However, Schwarz presents empirical arguments against this analysis(which was originally proposed by Heycock and Kroch) and favors a
gapping analysis with the nonfinite clause sich hinlegen as the sole rem-
nant. Actually, the only possible analysis on the model developed here is
in terms of gapping — or rather subgapping, since the second conjunct
includes an overt complement verb (see Section 5.2.2, in particular exam-
ples [23b]–[23c] and [24b]–[24c], for a definition and illustrations of sub-
gapping). The hypothesis that gapping originates in the conceptualization
stage of sentence production (or, at any rate, before linearization hastaken place) leaves open the possibility that direct object fronting is real-
ized in one conjunct only — here: in the anterior conjunct. FCR is ruled
out because, due to obligatory left-peripheral borrowing, it would entail
borrowing the entire left-peripheral string of constituents and force Die
Suppe into the role of hinlegen’s direct object. An SGF analysis fails for
the simple reason that the posterior conjunct does not contain a finite
verb. (Variant [41c] is an SGF coordination.)
Schwarz (1998) also discusses variant (41d) (his example [41c]). Inorder to understand its ill-formedness, first consider (41f ), which em-
bodies BCR and gapping simultaneously (cf. [41g]). Gapping succeeds
here because each conjunct contains one member of each contrastive
Clausal coordination and coordinative ellipsis 685
pairs (die Suppe–das Bier and essen–trinken). The ill-formedness of (41d)
is due to the fact that the fronted direct object die Suppe is not properly
paired with a contrasting constituent but with the reflexive pronoun sich
‘himself ’, as indicated in (41e). Unlike die Suppe, however, sich is not al-
lowed to occupy the preverbal position in a main clause. In (41h), each of
the two complement clauses die Suppe zu essen and sich hinzuliegen is a
continuous constituent. The adjacency of these clauses allows a nonellip-tical analysis with the left-branching complement of versuchen as coordi-
nation domain.
(41) a. Die Suppe wird der Hans essen undthe soup will the Hans eat and
der Hans wird sich hinlegen
himself down-lie
‘The soup, Hans will eat and lie down’
(Schwarz 1998: example [1])
b. [Die Suppe]1 wird der Hans [VP t1 essen] und [VP sich
hinlegen]]
c. Die Suppe ißt der Hans und legt sich hin
the soup eats the Hans and lies himself down
‘The soup, Hans eats and he lies down’
d. *Die Suppe soll der Hans zu essen und sich
the soup should the Hans to eat and himself
hinzulegen versuchen
down-to-lie try
‘The soup, Hans should try to eat and to lie down’
(Schwarz 1998: example [41c])e. *Die Suppe soll der Hans zu essen versuchen und sich soll der
Hans hinzulegen versuchen
f. Die Suppe soll der Hans zu essen und das
the soup should the Hans to eat and the
Bier zu trinken versuchen
beer to drink try
‘The soup, Hans should try to eat and to drink the beer’
g. Die Suppe soll der Hans zu essen versuchen und das Bier soll
der Hans zu trinken versuchen
h. Der Hans soll die Suppe zu essen und sich hinzulegen versuchen
The assumptions made so far also explain the grammaticality contrast be-
tween (42a) and (42b), which di¤er only with respect to the presence orabsence of direct object fronting. Variant (42a) is ungrammatical for ex-
actly the same reason as (41d) — compare (41e) with (42b). The gramma-
ticality of version (42c) corresponds to that of (41h).
686 G. Kempen
(42) a. *Die Suppe lehnt der Hans zu essen und sich
the soup turns the Hans to eat and himself
hinzulegen ab
down-to-lie down
‘The soup, Hans refuses to eat and to lie down’
(Schwarz 1998: example [46a])
b. *Die Suppe lehnt der Hans zu essen ab und sich lehnt der Hans
hinzulegen ab
c. Der Hans lehnt die Suppe zu essen und sich
the Hans turns the soup to eat and himself
hinzulegen ab
down-to-lie down
‘The soup, Hans refuses to eat and to lie down’
(Schwarz 1998: example [47a])
In describing the incremental updating model I have assumed that it ap-
plies to Dutch and German alike. This does not necessarily imply that
Dutch examples elicit the same grammaticality judgments as their Ger-
man equivalents, and vice-versa. There may be factors outside of the
coordination and ellipsis domain that cause the Dutch and the German
acceptability judgments to diverge. A case in point is the greater word
order freedom in German than in Dutch. For instance, fronting a non-
pronominal direct object NP requires sharper focus and/or stronger con-trastive stress in Dutch than in German. Lack of this may be responsible
for the fact that (43a), although an acceptable equivalent of (41a), does
not sound as good as (43b). Another possible cause of Dutch-German
discrepancies may be di¤erential preferences for alternative elliptical
mechanisms. For instance, some speakers of German do not accept FCR
sentence (44a) but strongly prefer SGF alternative (44b), whereas Dutch
speakers probably find the FCR version (45a) better than SGF version
(45b).
(43) a. De soep zal Hans eten en zich terugtrekken
the soup will Hans eat and himself retire
(in zijn slaapkamer)
(in his bedroom)‘Hans will eat the soup and retire (to his bedroom)’
b. Hans zal de soep eten en zich terugtrekken (in zijn slaapkamer)
(44) a. ?Wegen des Sturms bleiben wir zuhause und
due-to the storm stay we at-home and
verschieben wir unsere Reise
postpone we our trip
‘Due to the storm we stay home and postpone our trip’
Clausal coordination and coordinative ellipsis 687
b. Wegen des Sturms bleiben wir zuhause und verschieben unsere
Reise
(45) a. Vanwege de storm blijven we thuis en stellen we onze reis uit
b. Vanwege de storm blijven we thuis en stellen onze reis uit
8. Explanatory value of coordination as updating
After having have shown that the incremental updating approach to FCRand gapping enables adequate analyses at least of their basic elliptical
properties, I now go one step further and claim important explanatory
advantages for the new approach.
8.1. The coordinate structure constraint (CSC) with across-the-board
(ATB) extraction
Ross (1967) proposed this well-known rule pair, which I reproduce in (46)
in the formulation by Sag, Wasow and Bender (2003: 444).
(46) Coordinate Structure Constraint (CSC):
In a coordinate structure
(a) no conjunct can be a gap,
(b) nor can a gap be contained in a conjunct if its filler isoutside of that conjunct . . .
Across-the-board exception (ATB addendum to CSC):
. . . unless each conjunct properly contains a gap paired with the
same filler.
Here, the term ‘‘gap’’ refers to a structural position that is empty due to
its filler having moved out of the coordination domain (‘‘extraction’’).
Part (a) of the CSC (‘‘no conjunct can be a gap’’) prevents empty con-
juncts such as in (47). Part (b) serves to eliminate cases such as (48b),
where the direct object whose tax fills a gap in the posterior conjunct
from where it has been extracted, but not a gap in the anterior conjunct.
Similarly, (48c) is ruled out due to the fact that extraction of a constituent(fronting of who within the relative clause) has applied in the anterior but
not the posterior conjunct. Grammaticality is indeed restored if extraction
is applied ‘‘across the board’’, as in (48d), where the subject gaps in both
conjuncts are paired with the same filler who.10
(47) *What sofa will he put the chair between some table and —?
(48) a. The nurse polished her trombone and the plumber computed my
tax
688 G. Kempen
b. *Whose tax did the nurse polish her trombone and the plumber
compute —?
c. *The nurse who polished her trombone and the plumber
computed my tax was a blonde
d. Who polished her trombone and computed my tax?
The coordination-as-updating approach provides an immediate account
for the CSC/ATB constraint. Replacing some ‘‘old’’ constituent within a
structure by a ‘‘new’’ one can only count as an update if two conditions
are met:
– The new constituent should fulfill all morphosyntactic constraints
that the embedding structure (i.e., the structure outside of the co-
ordination domain) imposes on the attachment site of the old
constituent.
– The new constituent should be able to duplicate all and only the at-
tachments of the old constituent to the embedding structure.
The first condition is violated by example (48c): The posterior clause
(the plumber computed my tax) does not qualify as a relative clause. No-
tice, incidentally, that the morphosyntactic constraints mentioned in the
first condition have the embedding structure as source and the old/new
constituents as target. For instance, if the head verb of a clause wants its
subject to be an NP, then this holds for both the old and the new con-stituent. However, the head verb cannot impose its current value of the
number feature on the new subject constituent because the subject is the
source, not the target, of the subject-verb agreement constraint. Hence,
verb number may take on di¤erent values before and after the update.
The second condition ensures that the new constituent fills exactly the
same positions in the embedding syntactic structure as the old one. To
illustrate, example (47) violates the second condition because the poste-
rior conjunct (what sofa) fails to fill the position occupied by its anteriorcounterpart (some table) and, instead, occupies an ‘‘external’’ position,
that is, a position outside of the coordination domain. The second con-
dition also applies to parts of the old constituent that have moved out
and occupy an external position. Example (48b) is a case in point since
part of the new clause (the direct object whose tax) occupies an external
position that was not occupied by the old clause. In (48d), the poste-
rior clause satisfies both conditions as it represents an independent Wh-
question just like the anterior one, and FCR licenses borrowing theWh-phrase. Sentence (49a) is well-formed because the posterior clause is
of the same type as the anterior one and mimics the former’s Fronting
operation that moved the Direct Object (apple bagels) to an external
Clausal coordination and coordinative ellipsis 689
position. (The coordination domain consists of the coordinated finite
complement clauses.) Variant (49b), however, is ill-formed in the reading
indicated by the brackets because the posterior clause fails to apply
Fronting to the Direct Object. Consequently, the posterior conjunct does
not fill an external position occupied by the anterior conjunct. Inciden-
tally, (49b) is syntactically correct if the second conjunct is analyzed as
an update for the entire first conjunct, that is, as consisting of two inde-pendent main clauses that together form a coordination domain (see the
brackets in (49c]). Here, the Fronting operation has not moved the Direct
Object out of the coordination domain.
(49) a. Apple bagels, I can assure you [that Leslie likes —] and
[Sandy hates —]
b. *Apple bagels, I can assure you [that Leslie likes —] and
[Sandy hates lox]
c. [Apple bagels, I can assure you that Leslie likes] and [Sandy
hates lox]
In conclusion, the coordination-as-updating idea appears to entail the
joint e¤ects of CSC-plus-ATB without any further stipulation — thus, in
a sense, explaining its existence.
8.2. Three-dimensional (or parallel) trees
In the past twenty-five years, it has been suggested repeatedly that some-
thing like three-dimensional trees would be an attractive format for repre-
senting coordination and other paratactic constructions (e.g., Goodall
1987; Grootveld 1992; van Riemsdijk 1998; De Vries 2003). The consecu-
tive members of a coordination could receive adjacent positions in the
third dimension and be aligned precisely so as to bring out their structuralparallelism. On the model elaborated above, this comes as no surprise.
The third dimension originates from consecutive updating operation,
each new update giving rise to a new conjunct (to an extra line in an up-
date diagram). So, the attractiveness of three-dimensional trees is an im-
mediate consequence of viewing coordination as incremental updating.
At the same time, the new perspective does not inherit an important
criticism leveled at other three-dimensional representations of coordinate
structures. As far as I know, they all presuppose full parallelism of match-ing constituents in the conjuncts. As we have seen in Section 3, this pre-
supposition is contradicted by the gapping facts, as borne out by exam-
ples such as those in (18) and (35).
690 G. Kempen
9. Discussion
The incremental updating model for coordinate ellipsis put forward in
this article was inspired by certain similarities between coordination and
self-repair. At the end of this exercise it appears that this source of inspi-
ration was valuable insofar as yielding plausible accounts for two central
elliptical phenomena under scrutiny: Forward conjunction reduction(FCR) and the varieties of gapping. The model may also be said to ex-
plain the well-known coordinate structure constraint with across-the-
board application of movement rules (CSC/ATB). This, in turn, implies
that an important type of coordination, SGF, cannot be analyzed in
terms of incremental updating because it violates the CSC/ATB con-
straint. Instead, I argued that SGF is not an elliptical phenomenon at
all. The fourth phenomenon — backward conjunction reduction (BCR)
— could not be analyzed as a product of incremental updating. However,I showed that it shares key properties with an elliptical structure with a
wider scope than coordination: left deletion. Hence, strictly speaking BCR
does not qualify as an instance of coordinate ellipsis proper.
The four types of clausal coordinate ellipsis were argued to originate in
four di¤erent stages of the model of the speaker: SGF in the intending
stage (i.e., during the preparation of the communicative intention), gap-
ping in the conceptualization stage, FCR during grammatical encoding,
and BCR during phonological encoding.A key property of coordination-as-updating approach is that it can be
viewed as largely independent from the grammar formalism that gener-
ates noncoordinate structures. It postulates a sentence production mode
that, in response to revisions or extensions of the speaker’s communica-
tive intention, attempts to save on processing load by only generating
the novel parts of the modified communicative intention instead of re-
processing of the complete revised/extended intention. This entails a dou-
ble theoretical advantage: Coordination-as-updating can be adopted rela-tively easily by extant grammar formalisms of heterogeneous design; and
is does not increase the computational complexity of these formalisms per
se (Harbusch and Kempen 2006).
The empirical evidence adduced above in support of the proposed ac-
counts of the four types of clausal coordinate ellipsis is entirely based
on grammaticality judgments, i.e., linguistic competence. This raises the
question whether these types are the only ones that occur in linguistic per-
formance, i.e., in spoken and written language production. The TIGERtreebank for written German (Brants et al. 2002) provides excellent mate-
rials to begin to answer this question. From this corpus, Harbusch and
Kempen (2007) extracted all coordinations with und ‘and’ that contained
Clausal coordination and coordinative ellipsis 691
at least two clauses (about 7000 sentences in total). More than half of
these clausal und-coordinations (about 4000) exhibited some form of co-
ordinate ellipsis. It was not di‰cult to categorize each of them — partly
automatically, partly by hand — under one of the four ellipsis types (or a
combination: BCR joined with one of the three ‘‘forward’’ types). All
four types occurred in considerable numbers. Importantly, in only 1 per-
cent of these elliptical coordinations (about 40), the shape deviated fromthe possibilities allowed by the proposed rules. Many of them concern
FCR (slashes ‘‘//’’ in the anterior clause mark the end of the borrowable
left periphery):
– Nonmaximal reduction. The ellipsis is applied only to part of the
elidable string. In (50), FCR licenses elision of the verb war ‘was’
in the posterior conjunct, but this option remains unused. (See also
Note 7.)– Overreduction. In particular, a non-peripheral part of a major con-
stituent of the clause is elided. Two examples are (51) and (52)
where the head nouns of the Subject of the posterior conjunct are
elided, thereby violating the constraint that in FCR only entire ma-
jor constituents are elidable.
– Peripherality violations by ‘‘little words.’’ In about 10 FCR cases,
the third-person reflexive pronoun sich (‘himself, herself, them-
selves’) was located at the ‘‘wrong’’ side of the boundary betweenthe borrowable left-periphery and the nonborrowable remainder of
the first conjunct. In (53), sich is too late to be shared by the other
conjunct. In (54), it is too early: it could be shared by the second
conjunct but this does not need a reflexive pronoun. (Cases like
these are not restricted to sich: One example involved the demon-
strative pronoun dies ‘this’.)
The small number of deviations does not warrant ranging these sentencesunder a separate elliptical category, the more so because upon closer in-
spection the authors might have considered them erroneous and corrected
them.
(50) Der Lehrkurs bei ihm // war streng und war grundlich
‘The course at his place was stern and was solid’
(51) . . . wahrend bei der Sparkasse X // Gebuhren von 50 zu berappen
sind und bei der Bank Y sogar Gebuhren von 60 Mark zu berappen
sind
‘. . . whereas at Savings Bank X fees of 50 are to be coughed up
and at Bank Y even fees of 60 Mark have to be coughed up’
692 G. Kempen
(52) Dabei schrumpfte // der Auftragseingang aus dem Inland um drei
Prozent und der Auftragseingang aus dem Ausland um vier Prozent
‘Moreover, the number of domestic orders shrank with three and
the number of orders from abroad with four percent’
(53) . . . wahrend // 78 Prozent sich fur Bush und vier Prozent sich fur
Clinton aussprachen
‘. . . while 78 percent expressed themselves in favor of Bush andfour percent themselves for Clinton’
(54) . . . daß sich weiß // davon am besten abhebt und von den
Autofahrern am ehesten gesehen wird
‘. . . that [the color] white gives the better contrast (lit.:
distinguishes itself best) and can be seen faster by the drivers’
Important open questions concern the possibilities of extending the scope
of the incremental updating model in various directions. First and fore-
most, does it generalize to other languages than German and Dutch? Sec-
ond, can incremental updating account for the syntactic aspects of coor-
dination and ellipsis at the level of NPs, PPs, APs, etc.? Third, could itilluminate the treatment of constructions that are cognate with coordina-
tion, in particular comparatives (cf. Lechner 2004) and appositions (De
Vries 2006; Kraak and Klooster 1968: 260)? I leave it to future research
to provide the answers.
Received 5 July 2005 Max Planck Institute
Revised version received for Psycholinguistics, Nijmegen
8 September 2006 Cognitive Psychology Unit, Leiden University
Notes
* I am indebted to Karin Harbusch for discussing many topics raised here, for making
available the corpus data mentioned in Section 9, and for writing the ELLEIPO simu-
lation program (Harbusch and Kempen 2006). I also thank Fieny Reimann-Pijls for
introducing me to the puzzles of coordination and coordinate ellipsis twenty-five years
ago. Correspondence address: Max Planck Institute for Psycholinguistics, P.O. Box
310, 6500 AH Nijmegen, The Netherlands. E-mail: [email protected].
1. For early explorations of the idea that coordination and coordinate ellipsis can be ex-
plained in terms of an analogy with self-repair, and the implication that coordinate
structure is iterative rather than recursive, see Kempen and Hoenkamp (1987), Pijls
and Kempen (1986), and Kempen (1991). For a recent computer program that applies
all forms of coordinate ellipsis to non-elliptical clausal coordinations in a manner based
on the theory presented in the present article, see Harbusch and Kempen (2006, 2009).
2. I will not deal with coordinations of more than two conjoint clauses, on the assumption
that they can be handled by relatively straightforward extensions of the theory pre-
sented here, in combination with a solution for asyndeton. (But see Borsley 2005.)
Clausal coordination and coordinative ellipsis 693
3. Long-distance gapping (see Section 5.2.2) is an exception to this generalization: It is
sometimes allowed to cut into nonfinite complement clauses.
4. The terms ‘‘conceptual encoding’’ (or ‘‘conceptualizing’’) and ‘‘grammatical encoding’’
are explained in Section 3.
5. In certain types of infinitival clauses, the head verb is obligatorily preceded immedi-
ately by the equivalent of to (Du. te; Ger. zu). I assume here that this item forms part
of the head.
6. Particles as intended here are constitutive elements of particle verbs and (in German
and Dutch) of separable verbs. Sometimes they are members of semantically contras-
tive pairs, as in go up and go down, and thus may be remnants in gapping constructions
alongside other major constituents.
7. The sentences in (12) probably become less bad if pronounced with emphasis on both
tokens of vanmorgen. Furthermore, repetition of unstressed function words in the pos-
terior conjunct degrades the coordination to some extent but is not fatal. In (i), repeti-
tion of the auxiliary zal ‘will’ is dispreferred.
(i) Deze gast zal vanmiddag zijn ko¤ers pakken en (?zal)
this guest will this-afternoon his suitcases pack and will
vanavond vertrekken
tonight leave
‘This guest will pack his suitcases this afternoon and (will) leave tonight’
8. See Neijt (1979: 186–187) for examples similar to those in (26).
9. The analysis of SGF constructions in terms of ‘‘multiple predication’’ presupposes that
the initial constituent of the anterior clause cannot be borrowed. Indeed, in (36a), the
modifier gister ‘yesterday’ in the anterior clause does not clash with vanmorgen ‘this
morning’ in the posterior conjunct. My conjecture is that the posterior SGF conjunct
cannot borrow anything at all.
10. Example (47) is identical to 2.18 in Ross (1967); those in (48) are based on 4.81 and
4.82 in Ross (1967). The examples in (49) stem from Sag et al. 2003: 443–444.
References
Beaver, John & Ivan A. Sag. 2004. Coordinate ellipsis and apparent non-constituent coordi-
nation. In Muller, Stefan (ed.), Proceedings of the HPSG04 Conference, Center for Com-
putational Linguistics, Katholieke Universiteit Leuven, August 3–6, 2004, 48–69. Stanford
CA: CSLI. http://csli-publications.stanford.edu/HPSG/5/ (accessed 23 February 2009).
Brants, Sabine, Stefanie Dipper, Peter Eisenberg, Silvia Hansen-Schirra, Esther Konig,
Wolfgang Lezius, Christian Rohrer, George Smith & Hans Uszkoreit. 2004. TIGER: Lin-
guistic interpretation of a German corpus. Research on Language and Computation 2.
597–620.
Buring, Daniel & Katharina Hartmann. 1998. Asymmetrische Koordination. Linguistische
Berichte 174. 172–201.
De Vries, Mark. 2003. Three-dimensional grammar. In Linguistics in The Netherlands 2003.
201–212.
De Vries, Mark. 2006. The syntax of appositive relativization: On specifying coordination,
false free relatives, and promotion. Linguistic Inquiry 37. 229–270.
Dik, Simon. 1968. Coordination: Its implications for the theory of general linguistics. Amster-
dam: North-Holland.
694 G. Kempen
Frank, Anette. 2002. A (discourse) functional analysis of asymmetric coordination. In
Miriam Butt & Tracy Holloway King (eds.) Proceedings of the LFG’02 Conference, Na-
tional Technical University of Athens, 174–196. Stanford, CA: CSLI Online Publications.
http://csli-publications.stanford.edu/LFG/7/lfg02.pdf (accessed 23 February 2009).
Goodall, Grant. 1987. Parallel structures in syntax: Coordination, causatives, and restructur-
ing. Cambridge: Cambridge University Press.
Grootveld, Marjan. 1992. On the representation of coordination. Linguistics in The Nether-
lands 1992. 61–73.
Haeseryn, Walter, Kirsten Romijn, Guido Geerts, Jaap De Rooij & Maarten C. Van den
Toorn. 1997. Algemene Nederlandse Spraakkunst. Groningen: Martinus Nijho¤.
Harbusch, Karin & Gerard Kempen. 2002. A quantitative model of word order and move-
ment in English, Dutch and German complement constructions. Proceedings of the 19th
International Conference on Computational Linguistics (COLING 2002), Taipei (Taiwan).
San Francisco: Morgan Kaufmann. http://www.aclweb.org/anthology/C/C02/C02-1034.
pdf (accessed 31 March 2009).
Harbusch, Karin & Gerard Kempen. 2006. ELLEIPO: A module that computes coordina-
tive ellipsis for language generators that don’t. In EACL-2006: 11th Conference of the
European Chapter of the Association for Computational Linguistics (Trento, Italy; April
2006). http://aclweb.org/anthology/E/E06/E06-2008.pdf (accessed 31 March 2009).
Haspelmath, Martin (ed.). 2004. Coordinating constructions. Amsterdam: John Benjamins.
Hendriks, Petra A. 2004. Coherence relations, ellipsis and contrastive topics. Journal of Se-
mantics 21. 133–153.
Heycock, Caroline & Anthony Kroch. 1994. Verb movement and coordination in a dynamic
theory of licensing. The Linguistic Review 11. 257–283.
Hohle, Tilman. 1983. Subjektlucken in Koordinationen. Unpublished manuscript, Tubingen
University. http://www.linguistik.uni-tuebingen.de/hoehle/manuskripte/SLF-W5.1_neu.
pdf (accessed 31 March 2009).
Hudson, Richard A. 1976. Conjunction reduction, gapping, and right-node raising. Lan-
guage 52. 535–562.
Johannessen, Janne B. 1998. Coordination. Oxford: Oxford University Press.
Johnson, Kyle. 2002. Restoring exotic coordinations to normalcy. Linguistic Inquiry 33.
97–156.
Kathol, Andreas. 1999. Linearization vs. phrase structure in German coordination construc-
tions. Cognitive Linguistics 10. 303–342.
Kathol, Andreas. 2000. Linear syntax. Oxford: Oxford University Press.
Kempen, Gerard. 1977. Conceptualizing and formulating in sentence production. In S.
Rosenberg (ed.), Sentence production: Developments in research and theory, 259–274. Hills-
dale, NJ: Erlbaum.
Kempen, Gerard. 1987. A framework for incremental syntactic tree formation. Proceedings
of the Tenth International Joint Conference on Artificial Intelligence (IJCAI’87). http://
dli.iiit.ac.in/ijcai/IJCAI-87-VOL2/PDF/011.pdf (accessed 31 March 2009).
Kempen, Gerard. 1991. Conjunction reduction and gapping in clause-level coordination: an
inheritance-based approach. Computational Intelligence 7. 357–360.
Kempen, Gerard & Eduard Hoenkamp. 1987. An incremental procedural grammar for sen-
tence formulation. Cognitive Science 11. 201–258.
Kempen, Gerard & Pieter Huijbers. 1983. The lexicalization process in sentence production
and naming: indirect election of words. Cognition 14. 185–209.
Kraak, Remmert & Wim Klooster. 1968. Syntaxis. Culemborg: Stam-Kemperman.
Lechner, Winfried. 2004. Ellipsis in comparatives. Berlin & New York: Mouton de Gruyter.
Levelt, Willem J. M. 1983. Monitoring and self-repair in speech. Cognition 14. 41–104.
Clausal coordination and coordinative ellipsis 695
Levelt, Willem J. M. 1989. Speaking: From intention to articulation. Cambridge, MA: MIT
Press.
Neijt, Anneke. 1980 [1979]. Gapping: A contribution to sentence grammar, 2nd edn. Dor-
drecht: Foris.
Pijls, Fieny & Gerard Kempen. 1986. Een psycholinguıstisch model voor syntactische sa-
mentrekking. De Nieuwe Taalgid 79. 217–234.
Postal, Paul. 1974. On raising: one rule of English grammar and its theoretical implications.
Cambridge, MA: MIT Press.
Reich, Ingo. 2008. From discourse to ‘‘odd coordinations’’: On asymmetric coordination
and subject gaps in German. In Cathrine Fabricius-Hansen & Wiebke Ramm (eds.),
‘Subordination’ versus ‘coordination’ in sentence and text: A cross-linguistic perspective,
281–303. Amsterdam: Benjamins.
Ross, John R. 1986 [1967]. Infinite syntax! (Language and Being). Norwood, NJ: Ablex.
[Constraints on variables in syntax. Cambridge, MA: MIT dissertation.]
Sag, Ivan A., Thomas Wasow & Emily M. Bender. 2003. Syntactic theory: A formal intro-
duction. Stanford, CA: CSLI.
Schwabe, Kerstin & Ning Zhang (eds.). 2000. Ellipsis in conjunction. Tubingen: Niemeyer.
Schwarz, Bernhard. 1998. On odd coordinations in German. Journal of Comparative Ger-
manic Linguistics 2. 191–219.
Seuren, Pieter A. M. 1996. Semantic syntax. Oxford: Blackwell.
Slobin, Dan I. 1996. From ‘‘thought and language’’ to ‘‘thinking for speaking.’’ In John J.
Gumperz & Stephen C. Levinson (eds.), Rethinking linguistic relativity, 70–96. Cam-
bridge: Cambridge University Press.
Steedman, Mark. 2000. The syntactic process. Cambridge, MA: MIT Press.
Sturm, Arie. 1995. SGF-coordinatie bij nader inzien: verklaring van de eigenschappen en on-
gehoorde mogelijkheden. De Nieuwe Taalgids 88. 333–353.
te Velde, John. 2006. Deriving coordinate symmetries: A phase-based approach integrating Se-
lect, Merge, Copy and Match. Amsterdam: John Benjamins.
Vallduvı, Enric & Elisabet Engdahl (1996). The linguistic realization of information packag-
ing. Linguistics 34. 459–519.
Van Oirsouw, Robert R. 1987. The syntax of coordination. London: Croom Helm.
van Riemsdijk, Henk. 1998. Trees and Scions — Science and trees. In Chomsky 70th
Birthday Celebration Fest-Web-Page. http://cognet.mit.edu/Books/celebration/essays/
riemsdyk.html (accessed 23 February 2009).
Van Wijk, Carel & Gerard Kempen. 1987. A dual system for producing self-repairs in spon-
taneous speech: evidence from experimentally elicited corrections. Cognitive Psychology
19. 403–440.
696 G. Kempen